Start-Up Commercialisation Strategy and Innovative Dynamics

Transcription

1 Unversty of Toronto From the SelectedWorks of Joshua S Gans October, 2007 Start-U Commercalsaton Strategy and Innovatve Dynamcs Joshua S Gans Avalable at: htts://works.beress.com/joshuagans/5/

2 Start-U Commercalsaton Strategy and Innovatve Dynamcs * by Joshua S. Gans Unversty of Melbourne Frst Draft: 15 th December, 2005 Ths Verson: 22 nd October, 2007 Ths aer endogenses a start-u s choce between comettve and cooeratve commercalsaton n a dynamc envronment. It s demonstrated that, deendng uon frms dynamc caabltes, there may or may not be gans to trade between ncumbents and start-us n a cumulatve nnovaton envronment; that s, start-us may not be adequately comensated for losses n future nnovatve otental. Because of ths, there s no clear relatonsh between observed nter-ndustry nnovaton and commercalsaton choce unless dynamc caabltes of frms are taken nto account. In addton, the analyss demonstrates subtle and novel nsghts nto the relatonsh between dynamc caabltes and rates of nnovaton. Journal of conomc Lterature Classfcaton Number: O31. Keywords. nnovaton, ncumbency, dynamc caabltes, lcensng, commercalsaton. * I thank Nsvan rkal, Thomas Hellmann, Yoss Segel, Scott Stern, and artcants at the NBR ntrereneursh Strategy and Structure Conference (Jackson, WY, 2007), the Australasan conometrc Socety Meetngs (Alce Srngs, 2006), 2007 IIO Conference (Savannah, GA) and Unversty of Vrgna for helful dscussons and the Intellectual Proerty Research Insttute of Australa (IPRIA) and an ARC Dscovery Grant for fnancal assstance. Resonsblty for all errors les wth me. All corresondence to Joshua Gans, Melbourne Busness School, 200 Lecester Street, Carlton VIC 3053 Australa, -mal: J.Gans@unmelb.edu.au. The latest verson of ths aer s avalable at

3 ntrereneurs rarely contnue to manage the frms they found; extng through acquston or erhas through on IPO that dlutes ther control. As s well known, these modes of ext have very dfferent roertes. Secfcally, start-us mght comete head-to-head wth ncumbents n roduct markets wth an IPO cementng that ndeendence. Alternatvely, they mght cooerate wth them through lcensng, allances or acquston actvty n markets for deas. 1 Teece (1987) emhassed that cooeraton was mutually benefcal for ncumbents and start-us as t avoded the dulcaton of crtcal comlementary assets that would be requred when start-us chose to enter roduct markets. Gans and Stern (2000) bult on ths and exlored the way n whch lcensng mght avod the dssaton of monooly rents but also how ntellectual roerty rotecton mght facltate such cooeratve commercalsaton by removng exroraton rsk that mght otherwse cause start-us to avod drect negotatons wth ncumbents (see also Anton and Yao, 1994; Arora, 1995). These drvers of cooeratve commercalsaton that s, the mortance of comlementary assets and the strength of ntellectual roerty rotecton were borne out n emrcal nvestgatons by Gans, Hsu and Stern (2002), Arora and Ceccagnol (2005) and Hsu (2006) who also consdered the mortance of ntermedares n markets for deas as facltators of cooeraton across ndustres. Ths lterature suggests two key conclusons regardng start-u commercalsaton and nnovaton. Frst, that cooeratve commercalsaton and oortuntes for t wll ncrease startu nnovatve ncentves. Put smly, the returns that can be acheved through cooeratve commercalsaton factor nto account a start-u s otental return from cometton (as ths s always an alternatve for them). Thus, cooeratve commercalsaton choces wll be correlated wth hgher nnovatve ncentves for start-u frms. An mlcaton of ths s that the share of 1 See Gans and Stern (2003) for a revew.

4 2 start-u or entrant nnovaton n an ndustry wll be hgher where cooeratve commercalsaton s more commonly observed. Second, that there are overwhelmng reasons why gans from trade n markets for deas should be ostve and that the man reason such trade s not observed s some falure n that market. Market falure could arse from weakness of roerty rghts, nformaton asymmetry medng effcent barganng or hgh search costs for cooeratve artners (Gans and Stern, 2003). These cause cooeratve agreements not be realsed and cometton to occur by default. For the most art, these key conclusons reflect a statc ntuton and otentally neglect mortant dynamc consderatons. For examle, many teachng cases examnng start-u commercalsaton choces hghlght mortant nternal debates regardng the mmedate gans from lcensng or acquston versus the concern that the start-u mght be sellng out too early and losng ther brthrght to future nnovatve returns (Bartlett, 1983; Cae, 1999) or otherwse mortgagng away ther comany s future (Psano, 1994,.10). These cases hyothesse that, when a start-u frm has oortuntes for develong nnovatons n the future, by cooeratng wth ncumbents today, those oortuntes are otentally dmnshed. Indeed, the artcants n the cases aear to suggest that ths future cost may be so great that a start-u frm should consder avodng lcensng and cooeraton altogether. To an economst, such ossbltes are generally seen as beng another factor n the rce of cooeraton: namely, a start-u frm would have to be comensated for any reducton n ts ablty to nnovatve n the future. However, even recognsng ths, t may be that the gans to trade between establshed frms and start-us are low or negatve when the mact of future nnovatve cometton s taken nto account. Therefore, to roerly take these consderatons nto account requres a dynamc model; the rovson of whch s the contrbuton of ths aer.

5 3 To acheve ths, I buld uon a tractable framework for exlorng start-u or entrant nnovaton n the context of comettve nteractons wth an ncumbent frm that was develoed by Segal and Whnston (2007) hereafter, SW. SW only examned a lmted form of lcensng and only consdered cometton and the effect of ncumbent anttrust ractces on rates of nnovaton. 2 In addton, SW assumed that the same frms would ersst n the ndustry through successve waves of nnovaton; somethng I relax here. Secfcally, the model set-u here consders an envronment where at any gven ont n tme there are (effectvely) at most two frms n the ndustry an ncumbent and an entrant. 3 As n SW, an entrant today may become an ncumbent tomorrow and vce versa. As the focus here s on start-u commercalsaton choces and nnovatve ncentves, I focus on a model where the start-u s choce of nnovaton s endogenous. However, unlke SW, I also allow ncumbents to assume an nnovaton leadersh role (although ther nnovaton ntensty s treated as exogenous for tractablty reasons). When an entrant nnovates, f there s no cooeraton (.e., lcensng or acquston), t dslaces the ncumbent for the next generaton of nnovaton. If there s cooeraton, the ncumbent s not dslaced and reserves ts role. A key set of arameters n the model consders the dynamc caabltes of the frms. SW assumed that, should an ncumbent be dslaced, then t, wth certanty, becomes the entrant for the next generaton. Ths can be nterreted as a strong form of dynamc R&D caabltes whereby a current ncumbent has a sgnfcant advantage as an nnovator towards the next roduct generaton n that t reemts others from contestng the nnovaton market. Here I relax ths assumton by allowng that ncumbent caablty to range from non- 2 Other work on cumulatve nnovaton smlarly does not endogense the commercalsaton choces of start-us (see, for examle, the survey by Scotchmer, 2005). 3 In actualty, the model exlctly allows for many frms and ths s crtcal to the analyss and conclusons. However, through smlfyng assumtons I derve a stuaton where consderaton s requred of only two frms at any gven stage of the dynamc game.

6 4 exstent (the ncumbent cannot engage n future nnovaton at all) to strong as assumed by SW. In addton, because my model consders lcensng whereby the ncumbent s not dslaced, I also consder entrant dynamc caabltes. That s, should an entrant lcense ts nnovaton, there s some robablty that t wll reemt others n the nnovaton market. If ths robablty s hgh, the entrant s sad to have strong dynamc caabltes. However, a case that wll be of nterest s where ths robablty s low and cooeraton results n low rosectve returns for the start-u from future nnovaton. It s ths ossblty that ermts the dynamc analyss of notons that lcensng or acquston may nvolve start-us sellng ther brthrght to future nnovatve rents. In essence, I allow the commercalsaton decson to mact both on the tye of cometton n roduct markets today as well as the structure of cometton n nnovaton markets n the future. Ths hghlghts a dual mact from commercalsaton choces. Ths modfcaton reflects realty. Secfcally, there are many nstances where future nnovatve otental rests wth those who have nnovated n the resent. For nstance, Nklas Zennstrom and Janus Frs founded the eer-to-eer fle sharng network, KaZaA, whch was acqured by Sharman, before movng onto found the eer-to-eer IP telehony network, Skye, whch tself was acqured by ebay. They have now moved nto IP televson wth a new venture, Joost. In each case, they have leveraged sklls to become a lead nnovator n the next generaton of eer and fast transfer Internet technologes. In other cases, the leverage of dynamc caabltes has led to drect cometton for the ntal venture. Steve Jobs founded Ale n the 1970s but left n 1986 followng dsagreements on frm drecton to found NeXT and Pxar. Ten years later NeXT was acqured by Ale wth ts oeratng system to become the core of the hghly successful OSX. Pxar was acqured by Dsney n Smlarly, Walt Dsney, havng been rebuffed and seen hs anmaton deas

7 5 exrorated by several studos, went on to found hs own comany and domnate the entre ndustry (Gabler, 2006). In contrast to Jobs (whose technologes and sklls was acqured), Dsney was to use hs dynamc caabltes to take on establshed frms n the roduct market and hmself become the market leader. Wth ths framework I fnd that some mortant and subtle, dynamc effects that sgnfcantly qualfy the ntuton of statc models of nnovaton. Frst, the returns from lcensng are drven by mmedate savngs (avodng dulcaton of comlementary assets and dssaton of monooly rents) but also by the value of ncumbent technologcal leadersh. That value s tself endogenous n a dynamc envronment and t s demonstrated that t can be sometmes lower under lcensng than under cometton; mtgatng start-u nnovatve ncentves n equlbrum. Nonetheless, t s demonstrated that when lcensng s an equlbrum outcome, start-u nnovaton rates are hgher when lcensng s ermtted than when t s not; confrmng standard ntuton. However, a key fndng here s that the gans from trade from lcensng may not always be ostve. In a stuaton where the dynamc caabltes are very asymmetrc, lcensng means that some future nnovatve rents are jontly forgone by the current ncumbent and entrant n favour of future entrants. In contrast, cometton means that such rents (even f they are lower) are catured by current frms as the entrant becomes the ncumbent and the ncumbent becomes the next entrant. Thus, deendng uon the relatve dynamc caabltes, both frms may fnd ths mutually referable to cooeratve commercalsaton. Ths catures some of the case-based ntuton that dynamc caabltes may favour contnued cometton but also hghlghts some subtletes n how such caabltes generate ths outcome. The mortant mlcaton of ths s that there s no smle relatonsh between rates of

8 6 nnovaton and commercalsaton strategy because the latter tself deends on ndustry and frm characterstcs. Thus, emrcal researchers need to consder the nature of dynamc caabltes as mortant controls n understandng the assocaton between nnovaton levels and commercalsaton choce. The aer roceeds as follows: n Sectons 1 and 2, the basc model s ntroduced and the equlbrum under no lcensng (or cometton) s resented. Secton 3 then consders the lcensng case ncludng a dervaton of the lcensng fee n a dynamc context. Imortantly, ths demonstrates that ncumbency advantage even f not forfeted n equlbrum does mact on nnovaton benefts n ths case and characterses the gans from trade from lcensng. Secton 4 then analyses acquston as oosed to lcensng as a form of cooeratve commercalsaton. It s demonstrated that these modes have dstnct dynamc dfferences. Hghlghtng those s a searate contrbuton of the aer. Secton 5 then consders the mact of lcensng on equlbrum nnovaton rates and also the relatonsh between those rates and dynamc caabltes. A fnal secton concludes. 1. Model Set-U The basc set-u of the model follows SW but wth an mortant generalsaton. Whereas SW assume that a dslaced ncumbent becomes an nnovatng entrant wth certanty, here I allow for a more comettve structure to entrant nnovatve actvty. As I descrbe below, t s no longer certan that a dslaced ncumbent wll become the nnovatng entrant and, also, as ths matters when consderng cooeratve commercalsaton, an entrant cannot be guaranteed a role as an nnovatng entrant n the future.

9 7 Frms and Innovatons The model nvolves dscrete tme and an nfnte horzon wth the common dscount rate for all artcants of [0,1]. Innovatons occur sequentally wth each nnovaton beng a new roduct that yelds valuable qualty advantages over the revous generaton. The frm that develos the nnovaton receves an nfntely lved atent on t; although the exected economc lfe of the roduct wll be fnte. At any gven ont n tme, there s one frm, the ncumbent (I), who holds the atent rghts to the current leadng roduct or generaton. Aart from the frst erod, the roduct generates a constant flow of monooly rents,, untl such tme as t s dslaced by a new nnovaton. Whle SW assumed that only an entrant frm would exend resources on nnovaton, here t s assumed that there s a ool of frms (nfnte n number) and ncludng the current ncumbent that could otentally engage n nnovatve actvty. When t s the current ncumbent, nnovaton for them carres an addtonal short-term roft reward,, that they earn n the erod they nnovate. I deart from SW by also allowng for the ossblty that t s the ncumbent and not an entrant who takes the leadersh n nnovaton. Followng O Donoghue, Scotchmer and Thsse (1998), I assume that only one of these (to be termed the entrant or f t s not the current ncumbent, I) s selected at random wth the caablty to nvest n research resources towards that next generaton. Ths frm s the nnovaton leader. One can concetualse ths stuaton as one where deas for the next roduct generaton occur at random and are granted to only one frm who then chooses the level of resources to nvest towards realsng t as a vable nnovaton. Therefore, for any gven frm, the robablty that they wll engage n nnovatve entry s nfntesmal. However, as I dscuss below, for exstng artcants n the ndustry, I

10 8 consder what haens when they have an advantage n beng selected as the nnovaton leader. As noted earler, ths wll be a dearture from SW s assumton where an ncumbent becomes the next nnovatve entrant wth certanty. In each erod, the nnovaton leader,, chooses ts R&D ntensty, [0,1] lterally the robablty that t generates an nnovaton n the current erod. c( ) s the cost of R&D where c(.) s a non-decreasng, strctly convex functon wth c(0) 0. In ths set-u, an nnovaton leader who s an entrant has an ncentve to engage n ostve R&D effort as ths can only lead to mroved rofts. However, for an ncumbent nnovaton leader ths may not be the case. Greater nnovatve effort may lead to a stuaton where ther future ncumbency s laced at rsk (somethng that wll become more exlct below). Clearly, f they set I 0, then, should an ncumbent become the nnovaton leader, no further nnovaton would occur n the ndustry as ther oston would be nfntely lved. Ths s an artfact of our assumton made for smlfcaton that there s only one nnovaton leader at any artcular moment. Clearly, f there was cometton amongst more than one leader or some robablty that the leader was dslaced, ths ssue would not arse. For smlcty, as the focus of ths aer s on start-u entrereneural nnovaton and choces, I wll adot an assumton that there s a mnmum level of R&D effort that an ncumbent nnovaton leader must exend n order to retan ther leadersh. Moreover, gven ths, I wll assume that condtons are such that (a) that mnmum effort bnds and so R&D effort s set at that ont; whch I wll contnue to notate as I and (b) that the contnuaton rofts of the ncumbent at ths mnmum effort are ostve (so a artcaton constrant s satsfed). Thus, for all ntents and uroses, ncumbent R&D effort, should t aly s treated exogenously. Ths assumton can be relaxed n varous ways that add notatonal comlexty but

11 9 no addtonal nsght wth resect to the man results of the aer. Commercalsaton Choces When a new roduct s generated by an entrant, the atent holder,, faces a choce. It can enter nto roducton of the roduct generaton (cometton) or t can negotate wth the current ncumbent (cooeraton). 4 Followng ths, Nature then decdes whether the frm that does not hold atent roducton rghts s selected amongst the ool of frms to become the next entrant. If chooses a comettve ath, I loses ts monooly rofts and for the next erod both 1 t and earn rofts of, where 2. 5 ntry nto the roduct market costs sunk exendtures of f. I assume that such entry s credble; f. As we wll see, ths creates a value for ncumbency that macts uon the nature of comettve dynamcs. Followng that, f another nnovaton has not occurred, assumes the role of I and earns a roft flow of. The revous ncumbent then becomes one n the ool of frms from whch the next entrant wll be selected. also has a chance of becomng the nnovaton leader but n the ncumbent role. For smlcty, t s assumed that ncumbent would have to snk costs, f, f t wshed to re-enter the roduct market wth a new future nnovaton. Alternatvely, f engages n cooeratve commercalsaton t negotates to sell I an exclusve lcense to ts nnovaton. 6 I assume that such negotatons take the Nash barganng 4 Ths s a common resumton n nnovatve ndustres; see Teece (1987). 5 SW assume that, n cometton, I and earn dfferent roft levels. Ths s mortant to ther analyss but as t lays no secal role here, to save notaton, I assume roft levels under cometton are symmetrc. 6 It s mlctly assumed that f were to engage n non-exclusve lcensng, then the resultng on-gong cometton between two frms n roduct markets would be so ntense as to make entry non-credble. Of course, lcensng terms can be utlsed to soften such cometton. In ths case, however, the roft macts of an exclusve and non-exclusve lcense would be the same.

12 10 form where the ncumbent and entrant both have equal barganng ower. 7 If a lcensng deal s successfully negotated, receves a once-off ayment,, whle I reserves ts monooly oston. A lcensng agreement avods the mmedate comettve erod between and I and also the need for to ncur entry costs of f. In ths stuaton, t s who returns to the ool of frms as a otental future entrant whle I has a chance of becomng the nnovaton leader as an ncumbent. Dynamc Caabltes A novel feature of the model here s that the set of nnovatng frms can change from generaton to generaton. Secfcally, I allow both for the ossblty that, followng a successful nnovaton, a frm s resent n the market durng the develoment of the next generaton and the ossblty that they are not. As noted earler, for most models of atent races and nnovaton, dslaced ncumbents ext the ndustry whle for SW a dslaced ncumbent merely forgoes technologcal leadersh; takng on the role of the entrant. Here I nest both of these ossbltes. Recall that, followng successful entrant nnovaton, the next nnovaton leader s selected from an nfnte ool of frms; ncludng the dslaced ncumbent, n the case of cometton, or the entrant, n the case of cooeraton. However, there are dstnct reasons why each mght have a greater chance of beng selected from that ool; that s, an advantage n future nnovatve cometton. For a revous ncumbent who s not an nnovaton leader, knowledge of the ndustry may afford them wth an advantage due to sueror knowledge of the market and customers. Ths s a caablty that arses as a result of beng a roducer. To cature ths, I assume that 7 In a non-cooeratve barganng model, Gans and Stern (2000) show that ths outcome s the uer bound on the entrant s barganng ower when IP rotecton s otentally weak and the ncumbent can nvest n work around technologes.

13 11 followng successful ast nnovaton n the ndustry, wth robablty [0,1], the ncumbent becomes the nnovaton leader for the next generaton (the subscrt here standng for nnovatve caabltes generated by vrtue of beng a roducer). Ths mght be as an ncumbent or entrant deendng uon whether cooeratve commercalsaton occurs or not. Otherwse, they (effectvely) ext the ndustry and another frm takes on the role of the entrant. For an entrant who ursues cooeratve commercalsaton, ther future nnovatve advantage may arse because of ther knowledge of the nnovatve rocess for ths lne of roducts. To cature ths, I assume that an entrant who nnovates, wth robablty [0,1] (the subscrt here standng for nnovatve caabltes generated by vrtue of beng an nnovator), becomes the nnovaton leader (agan as an ncumbent or entrant as the case may be). Otherwse, they ext and are, otentally, relaced by a new entrant. As noted earler, ths rovdes a means of arametersng and modelng an nnovator s brthrght to future nnovatve rents. It catures the start-u s advantage n generatng future nnovatons. Fnally, the revous ncumbent mght also be an nnovaton leader. In ths case, they combne the knowledge from roducton and nnovaton and ths translates nto a robablty, [0,1] that they wll contnue as the nnovaton leader for the next generaton (the subscrt here standng for caabltes generated by vrtue of beng both a roducer and an nnovator). Ths robablty can also arse f an nnovatng entrant and a non-nnovatng ncumbent were to ntegrate through an acquston (rather than lcensng). One nterretaton of these arameters s that a frm s lkely to transton between roduct generatons f t has a dynamc R&D caablty. A frm s caabltes are usually defned n terms of ther ablty to delver roducts of a certan qualty and at a certan cost. Ths ablty then defnes the oston wthn a comettve marketlace. Dynamc caabltes are a

14 12 ste beyond ths and refer to a frm s ablty to transton n a changng envronment. For nstance, Teece, Psano and Shuen (1997) defne dynamc caabltes as the frm s ablty to ntegrate, buld, and reconfgure nternal and external cometences to address radly changng envronments. (.516) In ths aer, I examne how nnovatve caabltes mact uon the level of nnovaton acheved n an ndustry. In ths resect, the model here examnes successve generatons of roducts n an ndustry where each generaton s arrval deends endogenously on the resources drected towards nnovaton and R&D actvtes. A lmted number of frms wll have a caablty to conduct nnovaton so as to develo the next roduct generaton. Those caabltes may come externally through entry. Alternatvely, they mght be develoed nternally by those who are currently nnovatng towards the next roduct generaton. In ths resect, a frm s sad to have a dynamc caablty f they are able to successfully engage n develoment of the roduct generatons beyond that beng develoed today. As wll be demonstrated below, ths set-u enables exloraton of the role of dynamc caabltes n the nnovaton rocess as well as consderng ndvdual frm ncentves to nvest n such caabltes. Nonetheless, ths s a hgh level analyss n that I do not exlore the sources of such caabltes nor take a vew on how they are mantaned (cf: Sutton, 2002) No Lcensng Case To rovde a ont of comarson wth SW, I begn wth the case where lcensng s not ossble. In the nfnte-horzon dynamc game, followng SW, I confne attenton to statonary Markov erfect equlbra usng SW s dynamc rogrammng aroach. For ths urose, let V I 8 It also does not take nto account that the caablty tself may be a functon of commercalsaton choces (e.g., that lcensng mght gve a start-u cash to fnance the next nnovaton generaton).

15 13 be the exected resent value of rofts of a non-nnovatng ncumbent frm at the begnnng of any gven erod, These values wll satsfy: V I those for an nnovatng ncumbent and V those of an nnovatng entrant. V (1 ) V ( f V (1 ) V ) c( ) (V) I I V (1 )( V ) ( V ) (VI) I I V (1 )( V ) ( V (1 ) V ) c( ) (VI-) I I I I I I I where s the R&D ntensty chosen by an nnovatng entrant and I s the R&D ntensty chosen by an nnovatng ncumbent. Note that, followng an entrant nnovaton, the entrant contnues n the ndustry by default (as the ncumbent) whle the ncumbent may only wth robablty contnue n the ndustry as an nnovatng entrant. In addton, wth robablty the new ncumbent becomes an nnovatng one. If an ncumbent generated the nnovaton, wth robablty t contnues as the nnovator for the next generaton. As noted earler, we treat ncumbent R&D ntensty as exogenous. For an entrant nnovator, the equlbrum level of R&D ntensty s gven by the followng set of equatons: [0,1] I I 0 arg max f V (1 ) V V c( ) Followng SW, we let W denote the nnovaton rze or beneft. In ths case, W f V (1 ) V V (IB-Com) I I so that an entrant s effectvely solvng n each erod: [0,1] W c arg max ( ) (IS) Gven the convexty of c(.), ths gves an nnovaton suly relatonsh between the quantty of R&D ( ) and ts rce (W). As we wll see, wth all cases consdered below, all that changes

16 14 s how W s determned whle the (IS) relatonsh tself s otherwse stable. The convexty of R&D costs means that s non-decreasng n W. The equlbrum level of R&D by (should they have the oortunty), s determned by solvng (VI), (VI-) and (V) smultaneously and usng these to fnd the ntersecton of the (IB) and (IS) functons. The IB equaton descrbes the nnovaton beneft relatonsh between R&D ntensty and the level of the nnovaton rze. Any level of R&D ntensty that jontly satsfes (IS) and (IB) s a statonary equlbrum of the R&D game. The exressons nvolved are not of drect nterest and so are stated n the Aendx. Fgure 1 dects the equlbrum outcome. 9 The equlbrum rate of nnovaton, ˆ, occurs where the (IS) and (IB) curves ntersect. 10 At ths stage, t s useful to note that the equlbrum level of R&D wll be non-decreasng n, non-decreasng n, non-decreasng n and non-ncreasng n f. 11 Bascally, the frst three changes cause the IB curve to shft outwards whle the remanng change causes t to shft nwards. The IS curve s unchanged by another of these arameters. The ntuton for the comaratve statc on s nterestng. The more lkely t s for the ncumbent to ersst n the ndustry, the hgher s V. So long as V 0, the ossblty of I ersstence can only add to ncumbent value. Smlarly, the rofts of both the entrant and the ncumbent are dscounted by ther lkelhood of ersstng as an ncumbent. Ths lkelhood s ncreasng n so V rses as well. So an ncrease n unambguously rases the 9 For convenence these are drawn as straght lnes. 10 The IB curve may not be monotonc. SW demonstrate, however, that the same qualtatve analyss holds whether t s monotonc or not. For that reason, I smlfy the grahcal exoston for the more famlar downward slong case. 11 Ths can be seen by takng the dervatve of W n (IB-Com) wth resect to each varable and alyng Theorem 1 of Mlgrom and Roberts (1994) on the corresondng set of equlbra.

17 15 ncumbency advantage ( VI V ) and hence the nnovaton beneft. Fgure One: qulbrum under Cometton W IS IB ˆ Ths means that, under cometton, the more lkely t s that an ncumbent has a caablty to nnovate f dslaced, the hgher the rate of nnovaton wll be n the ndustry. Whle t may aear at frst glance that ncumbents nterested n slowng the rate of nnovaton so as to reserve ther ncumbency for longer may not wsh ths to haen, when they are the ncumbent, t s better to have a caablty than not. If t had to nvest n that caablty n each erod, t would be wllng to ay u to V for the oton. Note also that ths ncentve s even stronger f the entrant exects to make such an nvestment should t become the ncumbent (as ths has the ndrect effect of ncreasng ther rate of nnovaton).

18 16 3. Lcensng and Cooeraton I now turn to consder lcensng. 12 I wll contnue to assume here that entry s credble, f. In ths case, n negotatons wth a atent holder, the ncumbent earns V (1 ) V from lcensng for a fee of but otherwse exects to earn I I V (as entry occurs and ncumbency s lost). The nnovator exects to earn V from lcensng (as t may not ersst n the ndustry) and f V (1 ) V otherwse (as t gans, wth certanty, an ncumbency advantage from entry). There wll be gans to trade through lcensng f: I I VI (1 ) VI V V f VI (1 ) VI Jont Payoff from Cooeraton I I (2 f ) V V V Jont Payoff from Cometton (1) Note that cooeraton avods the dssaton of monooly rents and the sunk costs of entry; (2 f ). These are the same factors that drve the gans from trade for lcensng n statc models (see Salant, 1982; Gans and Stern, 2000). However, here there s also a dynamc comonent to the jont surlus from lcensng. Frst, a lcense agreement wll allow the ncumbent to reserve ts rofts and wll reclude an entrant from caturng those rofts. As there s only one ncumbency rent, ths nets out as a gan from trade from lcensng. Ths would also be the case for entrant rofts f the ncumbent and entrant had smlar robabltes of contnung as an nnovator. However, when these robabltes dffer, lcensng generates a gan to jont surlus of ( ) VI V VI. Notce that ths may not be a gan 12 As noted n the ntroducton, lcensng s only one form of cooeratve commercalsaton. It fts the formal model and so I focus uon t here; commentng on dfferences wth other forms of cooeraton n the concludng secton.

19 17 at all f () the rofts of an ncumbent nnovator are low; and () the robablty that the entrant, as a lcensor, contnues to nnovate towards the next generaton s less than the robablty that a dslaced ncumbent can do so; that s, f. Ths mght occur f the entrant has a more secalsed focus on the current generaton whereas the ncumbent has caabltes that gve t an R&D advantage n the next generaton. Take the extreme case where 0 and 1 and where V I V. By sgnng a I lcensng agreement, nether arty earns V whle by not sgnng the ncumbent earns V. In effect, the lcensng agreement confers a ostve externalty on a thrd arty (a otental entrant) whch s nternalsed f no lcensng agreement s reached. Indeed, f V 2 f, the overall gans from trade from lcensng would not be ostve and lcensng would not occur. As the contnuaton ayoffs are endogenous, care must be taken to establsh the exstence of an equlbrum wth lcensng. Determnng the condtons under whch lcensng wll actually take lace n equlbrum nvolves dervng the equlbrum value of contnuaton values under lcensng whch tself requres a soluton for. Gven ths, I emloy the Nash barganng soluton to determne the lcense fee. Assumng for the moment, that the gans from trade are ostve, let [0,1] denote the barganng ower of the entrant. Then the lcense fee,, s found by solvng: 1 VI VI V V f VI VI max (1 ) ( (1 ) ) (2) Ths gves (2 f )(1 ) V ( (1 ) )( V V V ). I I I In the lcensng case, the (conjectured) equlbrum contnuaton ayoffs are:

20 V V ( V V ) (VI) I I I I V (1 ) V ( V ) c( ) (V) 18 whle the equaton for V I remans as n the cometton case. Notce that, along the (conjectured) equlbrum ath, ncumbency nvolves a contnual flow of monooly rofts () eered by the ayment of lcense fees to reserve technologcal (and market) leadersh. In contrast, otental entrant returns are governed by the erodc earnngs from lcense fees over the economc lfe of the atent. In ths case, the nnovaton rze s: W (1 ) V (3) Thus, as n the case of cometton, under cooeraton the (IB) curve ncludes a factor based on the value of ncumbency; though, even though ths s never lost n (the conjectured) equlbrum. Nevertheless, entrant nnovators can stll arorate art of ths n negotatons over the lcense fee. 13 The (IS) relatonsh remans the same as the no lcensng case. The aendx states the resultng equlbrum contnuaton values. We are now n a oston to characterse (artally) the equlbrum outcome. Prooston 1. Lcensng s an equlbrum outcome for suffcently small and/or 0. Lcensng s not an equlbrum outcome for suffcently large as () 1; or () 1. The roof s n the aendx. The rooston reflects what we found examnng condton (1) but reveals an addtonal factor. Two thngs generate a stuaton where lcensng s not an equlbrum. Frst, farsghtedness () whereby weght s laced on dynamc factors as oosed to statc ones. Second, caablty asymmetry whereby ether the ncumbent or the entrant has a 13 Of course, ths would not be ossble f roduct market entry were not credble. Note, however, ths does not requre the nnovator to exercse ths entry oton, merely to facltate t (see also Anton and Yao, 1994).

21 19 sgnfcantly large oortunty to be the lead nnovator for the next generaton (.e., and are sgnfcantly dfferent). Our analyss of (1) revealed only the role of a dynamc advantage to the ncumbent. Closer examnaton also reveals that lcensng s not an equlbrum when the entrant s dynamc advantage s large. Intutvely, how ths oerates deends uon whether V V V 0 or not. When I I s small, V I s relatvely small and so the value of one of them becomng an ncumbent lead nnovator s less than the sum of values from one of them becomng the entrant lead nnovator. If s low whle s hgh, those jont returns are mroved by not lcensng as ths maxmses the lkelhood that one of them (n ths case, the current ncumbent) becomes an entrant lead nnovator. In contrast, when s large, V I s relatvely large and so the goal would be to maxmse the chances of one of the artes becomng an ncumbent lead nnovator. Note lcensng wll not acheve ths when s hgh whle s low as t s referable for the start-u to become the ncumbent. At ths ont, t s nstructve to return to the nformal case-based argument that cooeratve commercalsaton may not be undertaken because the start-u nnovator cannot be comensated for a loss of future nnovatve rents. The argument s that, by lcensng, the startu forgoes the ncumbency oston and the advantages that yelds n future nnovatve cometton. In our formal model here, ths factor would be most salent when s hgh. When ths s the case, an entrant who forgoes lcensng has a good chance of becomng an ncumbent who s the nnovaton leader. However, Prooston 1 demonstrates that ths nformal argument does not drve cometton n equlbrum. Frst, t s not smly that s large but that s large relatve to

22 20. Second, f s relatvely large, by not lcensng, the ncumbent and entrant maxmse the chance of realsng the rofts from an ncumbent nnovaton leader. However, ths s only mutually valuable n a stuaton where V V V 0 ; so that ndustry rofts are maxmsed I I by ensurng that an ncumbent s the nnovaton leader. Prooston 1 tells us that when s relatvely hgh, havng an ncumbent as the nnovaton leader does n fact maxmse ndustry rofts. Thus, n combnaton, a relatvely hgh does lead to the artes avodng lcensng. But Prooston 1 also demonstrates that the ooste can be the case. In a stuaton where s relatvely large, t may not be mutually benefcal to agree to lcensng. In ths stuaton, V V V 0, and so the artes would refer a stuaton that leads to one of them I I becomng the next entrant. When s large, the ncumbent forgoes become an nnovaton leader by not strkng a deal wth the start-u. Between them there s no loss and gan. In fact, t s because lcensng, n ths case, benefts other otental entrants that t may not be worthwhle. 14 In stuatons where s hgh, lcensng means that, n the next round of the nnovaton game, the weaker comettor amongst I and enters that game. In contrast, wth no lcensng, t s the stronger comettor that enters. Hence, there s a net loss from I and to others f lcensng occurs. Thus, clams for comettve commercalsaton are suorted but for dfferent and more subtle reasons than s artculated n case analyses. 4. Lcensng versus Acquston Lcensng s not the only form of cooeratve commercalsaton strategy. Another commonly ractced stuaton nvolves the start-u beng acqured by the ncumbent frm; 14 Ths also rases the ossblty that f there were multle ncumbents, a smlar external effect could arse. Multle ncumbents rase many other ssues and so ths ossblty s left for future research.

23 21 erhas n stuatons where a lcensng agreement or shft n ntellectual roerty rghts s nfeasble. The dfference between acquston and lcensng s that the start-u s removed from the ool of otental nnovators for the next generaton. However, ther caabltes are added to those of the ncumbents. Consequently, ths alters from to the chance that the ntegrated start-u wll become the nnovator for that generaton. Here, I consder when acquston mght be an equlbrum outcome relatve to cometton and also relatve to lcensng. There wll be gans from trade from acquston rather than cometton f: VI (1 ) VI V f VI (1 ) VI Jont Payoff from Cooeraton (2 f ) V ( ) ( V V ) I I Jont Payoff from Cometton (4) Note that the same statc gans from cooeraton (the LHS) occur for acquston as they do for lcensng. Where the two dffer s n the dynamc mlcatons. In artcular, an acquston causes the frms to jontly forgo a chance of earnng V but also otentally mroves ther chances of earnng V I rather than V I. Although ths wll only occur f. If, however, there are functonal dffcultes n ntegratng roductve and nnovatve caabltes, ths nequalty mght not hold; e.g., the nnovatve caabltes may be subsumed or extngushed n a larger organsaton. In that case, acquston reduces the lkelhood of the ncumbent becomng the lead nnovator n the next generaton. where, As for the (conjectured) equlbrum ayoffs n the acquston case, we have: V V ( V V ) (VI) I I I I V (1 ) V c( ) (V) V I s stll determned accordng to (VI-). The start-u s nnovaton rze s, therefore:

24 22 W V (IB-) Usng, the Nash barganng soluton, s gven by: (2 f )(1 ) V V ( (1 ) )( V V ). case. I I I Gven ths, we can rove a smlar rooston to art of Prooston 1 for the lcensng Prooston 2. Acquston s an equlbrum outcome for () suffcently small and/or () and suffcently small. Lke Prooston 1, acquston s an equlbrum when there s suffcent dscountng of the future. However, there are also some dfferences. In artcular, we cannot establsh suffcent condtons for acquston not to be an equlbrum even when 1 and the caabltes are chosen to maxmse the RHS of (4). In ths stuaton, we cannot rule out that (2 f ) V V V. I I Of course, n realty, frms may have otons of choosng between lcensng and acquston as a mode of cooeratve commercalsaton. Comarng (1) and (4), acquston wll have hgher gans from trade than lcensng f: V ( ) ( V V ) V V V I I I I ( )( V V ) V I I (5) The nterretaton here s qute ntutve. Acquston yelds the beneft of a otentally hgher robablty of ncumbent nnovaton leadersh wth the cost of losng a chance at an entrant oston n the next generaton. That these are the drvers s confrmed by the followng rooston: Prooston 3. Suose that suffcently small so that cometton s not an equlbrum outcome. Acquston s an equlbrum outcome as 0 and. Lcensng s an equlbrum outcome f.

25 23 The choce between acquston and lcensng would, n fact, hold for any. What ths mles s that, from Prooston 1, we know that lcensng s not an equlbrum outcome when, for examle, 1. In ths stuaton, so long as 1, lcensng would be chosen above acquston and so, by mlcaton, acquston s not an equlbrum outcome. Ths s consstent wth the ntuton gven by examnng (4) alone. It s worth emhaszng that Prooston 3 hghlghts a crtcal dfference between standard economc consderatons and a broader organzatonal or strategy vewont. In economcs, the rncle of selectve nterventon would suggest that addng the nnovatve caabltes of the entrereneur to an establshed frm would do no harm to the establshed frms rosects n becomng an nnovaton leader n the future. That s, there would be a resumton that. Note that n ths stuaton, acquston would be referred to lcensng. However, f, because of dffcultes n contractng over the leverage of dynamc caabltes, t was the case that, the frms may choosng a lcensng agreement allowng the entrereneur to return as an ndeendent nnovator rather than brngng the entrereneur nsde the frm. Ths s, ndeed, a common choce n cooeratve commercalsaton (Hellmann, 1996). The model here argues that ths can only be ratonalsed f, not only does the entrereneur add no value to an establshed frm but, n addton, causes that frm to be less lkely to be a future nnovator. 5. Innovaton Rate and Commercalsaton Strategy We are now n a oston to consder nnovaton rates across ndustres and ther

26 24 relatonsh to choces of cooeratve and comettve commercalsaton strateges. Not surrsngly, as that choce s tself contngent uon then nature of ncumbent and entrant dynamc caabltes, n the absence of a control for such caabltes, there may be no consstent relatonsh between commercalsaton strategy and the rate of nnovaton. To demonstrate ths, I wll roceed n stes. Frst, I wll examne the mact of lcensng on nnovaton rates controllng for caabltes. Second, I wll consder nnovaton rates as a functon of caabltes takng nto account the endogenous choce of commercalsaton strateges. In so dong, the am s to roduce a set of theoretcal redctons regardng the drvers of nnovaton across ndustres contngent uon varous controls emrcal researchers may have avalable. The mact of lcensng on nnovaton As the focus here s on start-u nnovaton choces, I adot a smlyng assumton here that I 0 and 0 ; that s, V I V. Here I wll also abstract away from dfferences n I caabltes and examne when haens to nnovaton rates as we move from a comettve to cooeratve commercalsaton attern. Ths can be seen as rovdng an analyss of nterndustry dfferences where factors other than caabltes (e.g., the strength of IP rotecton) account for varaton n commercalsaton strategy or alternatvely where caablty varatons can be otherwse taken nto account. As such, I wll assume, for the moment, that so as to smlfy notaton but also consder a stuaton where there are gans from trade from lcensng (as n Prooston 1). The framework rovded here rovdes a smle means of charactersng equlbrum rates of nnovaton. Gven the uward slong (IS) curve, that the rate of nnovaton wll be

27 25 hgher where the rze (W) s hgher. That s, lcensng wll result n a hgher nnovaton rate f W as defned by (IB-Coo) s hgher than that defned by (IB-Com) for any gven level of. In comarng (IB-Coo) wth (IB-Com), there are two factors to consder: the mmedate beneft from the current nnovaton and the on-gong beneft from an ncumbency advantage. Under lcensng, the mmedate beneft from the current nnovaton s (2 f )(1 ) whereas the returns from entry under no lcensng are f. Thus, an nnovator s able to gan more value under lcensng because t s able to threaten the ncumbent wth the loss of monooly rofts and arorate these wthout ncurrng sunk entry costs. In effect, the lcense fee s a share of the sum of ndustry rofts under monooly and those under cometton whereas under no lcensng the nnovator arorates ts ndvdual comettve rofts. As ndustry rofts under monooly exceed those followng cometton, the rze under lcensng s hgher. Counterng ths are future benefts. When, under both lcensng and no lcensng, the nnovator arorates the ncumbency advantage ( V I V ) less some share of the value from entry ( V ). However, for any gven, that advantage maybe greater under no lcensng than t s under lcensng. To see ths, note that: V I V 2 (1 )((1 ) f ) (1 )( f ) (1 (1 )) c( ) 2 (1 ) (2(1 ) (1 ) ) (IA-Com) V I V (1 ) ((1 )( f) ( )) (2(1 ) (1 ) ) (1 (1 )) c( ) (IA-Coo) 2 (1 ) (2(1 ) (1 ) ) It s easy to see that (IA-Com) wll exceed (IA-Coo) f: (1 ) (1 2 )(1 ) (6) 1 However, ths wll not hold for all arameter values but t wll hold f 2. When entrant

28 26 barganng ower s relatvely hgh, under cooeraton ther rofts are hgher whle the ncumbent s s lower. Ths reduces the ncumbency advantage. In effect, the returns to an entrant are hgher under lcensng and hence, the relatve value of ncumbency may be lower. The dscount rate weghts these two effects. Lookng at the overall rze, W, (IB-Coo) exceeds (IB-Com) f: ( 2 f ) (1 ) (1 ) 0 2 (1 ) (2(1 ) (1 ) ) (7) Not surrsngly, when there s comlete dscountng ( = 0), t s clear from (6) that W s hgher when lcensng occurs. When there s no dscountng, maxmum weght s laced on the ncumbency advantage n the rze. However, even n ths case, the dfference between (IB- Coo) and (IB-Com) becomes 0. The followng summarses ths result: 15 Prooston 4. When, the equlbrum nnovaton rate s hgher under lcensng than no lcensng. Bascally, the margnal mrovement n the ncumbency advantage under no lcensng never outweghs the mmedate benefts from lcensng the current nnovaton. Caabltes and nnovaton We now turn to consder the relatonsh between dynamc caabltes and the equlbrum rate of nnovaton. In Secton 3, t was noted that the equlbrum rate of nnovaton under cometton rose wth ; the robablty that a dslaced ncumbent would become an nnovatng entrant n the next generaton. By contrast, under lcensng, ncumbent dynamc caabltes are negatvely related to equlbrum nnovaton rates. Whle an ncrease n 15 The qualfer s made only for notatonal smlcty. Whenever, lcensng s an equlbrum outcome ths the rooston holds.

29 27 ncreases V I, t reduces the gans from trade through lcensng as t mroves the ncumbent s outsde alternatve. Ths has a drect and negatve mact on the lcense fee and overall a negatve mact on (IB-Coo). When the choce of commercalsaton strategy s endogenous, what then haens to nnovaton rate as the ncumbent caablty ( ) becomes stronger? As noted earler, ceters arbus, as rses the commercalsaton regme can swtch from lcensng to cometton. Over the range where lcensng s an equlbrum, nnovaton rates wll fall wth only to rse agan when cometton s the equlbrum outcome. Thus, there s a broad U-shaed relatonsh between ncumbent dynamc caabltes and nnovaton rates n an ndustry. Fgure 2 llustrates ths for our numercal soluton. In ths examle, the commercalsaton regme moves from lcensng to cometton as 0.3. Notce that the nnovaton rate declnes and then takes a dscrete downward jum as we move from lcensng to cometton. The nnovaton rate then rses agan and when 1 (for ths examle) actually exceeds the rate when 0.

30 28 2 Fgure Two ( c( ), 1, = 0.4, f 0.3, 0.5, 0.9, 0 ) 0.44 ˆ COMP ˆ COOP For entrants, dynamc caabltes only matter under lcensng. Under cometton, successful entrants become the ncumbent and so they automatcally ersst for the next wave of nnovaton. In contrast, should lcensng take lace, greater entrant caabltes mrove the robablty that the entrant s the leadng nnovator of the next of generaton (that s, Moreover, that robablty ostvely macts uon both the lkelhood that lcensng takes lace, the lcense fee negotated (as t s art of the entrant s outsde oton), and also the chances that successful nnovaton wll not drve the entrant from the ndustry. All of these combne to ncrease (IB-Coo) and so the equlbrum rate of nnovaton, under lcensng s ncreasng n. What ths means s that as ). rses, the commercalsaton regme may swtch from cometton to cooeraton and the equlbrum rate swtch from a constant level (unrelated to ) to a dscrete jum uwards and a steady rse thereafter. As such, n contrast to the stuaton wth ncumbent caabltes, there s a monotoncally ncreasng relatonsh between entrant

31 29 caabltes and the rate of nnovaton. 6. Concluson and Future Drectons Ths s the frst aer to consder the choce of start-u commercalsaton strategy n a dynamc envronment. It was demonstrated that dynamc consderatons mact uon ths decson n a way not catured by a urely statc focus. In artcular, the on-gong roles of the artes of a lcensng deal matter n terms of rent cature and the returns to lcensng over cometton. In turn, these on-gong roles are related to dynamc caabltes that s, the robablty that a frm wll have an nnovatve advantage n research towards the next generaton of roduct based on ts current role (as entrant or ncumbent). These caabltes feed back to determne the general relatonsh between commercalsaton actvtes of start-us and ther share of nnovaton across ndustres. In ths regard, erhas the most nterestng fndng was that start-us and ncumbents may not sgn cooeratve lcensng agreements even though ths would revent the dssaton of monooly rofts and dulcaton of comlementary nvestments. Ths occurred because to do so would send the start-u frm back to comete for the next generaton of nnovaton n stuatons where the ncumbent had stronger caabltes n ths regard. Ths naturally leads to the queston as to whether the frms could choose whch one of them would return to nnovatve cometton and whch would reman as the ncumbent. Ths s an nterestng ssue and n many resects goes to the heart of what a dynamc caablty s and how t s acqured. An ncumbent s lkely to be strong because of ts revous roduct market oston and ths lkely relates to nvestments t has made n the ast. An entrant would have to smlarly make those nvestments to strengthen ts future role and thus, one of the

32 30 gans from lcensng (reventng such dulcaton) would be lost. In addton, wth ant-trust laws, t s not clear that the ncumbent could cede ts roduct market oston so readly. Nonexclusve lcensng mght lay a role here but there would be some on-gong dssaton of monooly rents. Smlarly, the start-u could acqure the ncumbent. However, ths mght necessarly reclude t from becomng a strong nnovatve entrant unless some form of restructurng was ossble. Thus, there aears to be good reasons why changng ostons s not a smle choce and so t s natural to exlore nnovatve dynamcs when ths s mossble. However, a roer exloraton of these ssues remans an oen area for future research. There are several other drectons n whch the results of ths aer could be extended and exlored n future research. Frst, n ths aer, dynamc caabltes were consdered exogenously. ther frms had them (to a certan degree) or they dd not. In realty, the acquston of such caabltes s lkely to be a key and on-gong strategc choce for frms. Thus, endogensng ths choce and relatng those caabltes to more fundamental market condtons (as n Sutton, 2002) would aear to be a romsng avenue for future research. The model here rovdes a framework uon whch such an extenson mght be based. Fnally, ths model shares wth many others a smle consderaton of nnovatve strategy namely, nnovatve ntensty. Recent work by Adner and Zemsky (2005) goes beyond ths to consder macts on other strategc varables such as rces, market montorng, frm sze and the rate of overall technologcal rogress. Ther model s dynamc but does not consder the choce of commercalsaton strategy t only consders a comettve route for start-us. Lnkng ther aroach wth the endogenous choce of commercalsaton strategy as consdered here may lead to a rcher cture of the nnovaton envronment and the role of dslacng or dsrutve technologes on market and technologcal leadersh n an ndustry.